Research On Extracting Application Of Impervious Surface Information Based On Improved VGGNet

In recent years,with the development of remote sensing technology,the information recognition and extraction of impervious surface has important research significance and application value for urban land use change,urban ecological environment detection,urban hydrological change and urban planning.The impervious surface information is rich in content,the information is complex,and the interference of the atmosphere and the cloud layer during satellite shooting brings great trouble to the accurate identification and extraction of the impervious surface in the remote sensing image.Although breakthroughs have been made in the research of remote sensing in the field of remote sensing,some research results have been obtained on the identification and extraction of impervious surface information of remote sensing images at home and abroad.However,in the process of identification,most of them are manually manually Judging,greatly reducing work efficiency.With the accelerated construction of urbanization,the villages in the city are widespread,the messy characteristics of buildings and the obscuration of trees in the city have caused the accuracy of the impervious information extraction during satellite shooting to decrease.Therefore,this paper proposes a new method for impervious surface extraction based on VGGNet improved convolutional neural network for the low efficiency and low precision of the impervious surface recognition and extraction,and.through the encapsulation and secondary development based on PIE software.Apply the algorithm to the system.The main contents and results of this paper are as follows:(1)The remote sensing image impervious surface training data set is constructed for deep learning based on VGGNet convolutional neural network.Combined with the existing remote sensing image database,manually intercepting multi-spectral images of houses,roads,playgrounds and other categories,forming a set of 1150 sample sizes,of which the size is 256*256 fixed pixel size,of which 908 training samples,test sample 242 One.Take houses,roads,playgrounds,etc.as positive samples,and other water bodies,vegetation,bare land,etc.as negative samples.(2)A new method for impervious surface extraction based on improved VGGNet convolutional neural network is proposed.By changing the hierarchical structure of the model,and comparing the training effects of several common different convolutional neural network models on the impervious surface data set,Through the model structure adjustment,a convolutional neural network model with input layer->convolution layer->pooling layer->cavity convolution->anti-pooling is constructed.Experiments show that the improved model is relative to the original VGGNet.And the SVM method has the same improvement in the precision of impervious surface extraction,which not only solves the problem of low extraction precision,but also solves the problem of low labor efficiency.(3)Through the encapsulation of the improved model,combined with PIE software for secondary development,the design and implementation of the impervious surface information recognition and extraction system.